michaelttsai.com
A build-in-public site and living resume, hand-coded the second time around so I actually understand it.
What it is
A personal site where I document what I build and what each thing taught me. Three sections: About, Projects, Blog. No framework, no build step, no dependencies. Semantic HTML, one stylesheet, written by hand and deployed from a Git repository to GitHub Pages.
Why I built it
This is the second attempt. The first version I had an AI write end to end. It shipped in a weekend, it looked completely fine, and then I did not open it again for two months.
Working out why took longer than building it did. The honest answer is that I had no relationship with the code. Not one line of it was a decision I had made, so when I sat down to add something there was no thread to pull. I would have had to read my own website like a stranger before I could edit it, so I never did.
This version I am writing myself, with AI as a teacher instead of a ghostwriter. Slower to get live. Far more likely to still be alive in six months.
Outcome so far
- v1: built in a weekend, abandoned in two months, updated exactly zero times after launch.
- v2: live at michaelttsai.com as of July 2026, rebuilt from an empty folder. Every entry since has been added by me on its own branch and merged in.
The real measure is not whether it looks good on day one. It is whether there is a fourth entry, and a tenth, and whether the writing gets sharper as the record gets longer.
Lessons / what I'd do differently
- Shipping fast is worth nothing if you never come back. My first site cleared every bar except the only one that mattered, which was whether I would keep using it.
- Code you did not write is code you cannot maintain, at least not until you have read it as carefully as you would have written it. Using AI to skip the writing did not remove that cost. It moved it to a later date I had not planned for.
- Build the smallest version you understand completely, then add. v1 had features I never used, because I never got far enough inside it to use them.
- What I would keep: using AI aggressively, just pointed at explanation instead of output. Asking "why this tag and not that one" has taught me more in a week than the entire first build did.